196.81B / ~11B active (MoE)
Apache 2.0
StepFun flagship open foundation model positioned for frontier reasoning and agentic capabilities with high efficiency; sparse MoE architecture activates roughly 11B out of 196.81B total parameters per token; designed for AI agents, agentic coding, deep research, and tool orchestration workflows. Model card materials describe MTP-3 generation with typical 100-300 tok/s throughput (peaking around 350 tok/s in single-stream coding scenarios), 256K context via a 3:1 sliding-window/full-attention hybrid, and reasoning-mode control through chat-template configuration. Technical details highlight a 45-layer transformer backbone, 128,896-token vocabulary, and expert routing across 288 routed experts plus 1 shared expert with Top-8 activation per token. Reported benchmark highlights include AIME 2025, HMMT 2025, IMOAnswerBench, SWE-bench Verified (74.4), Terminal-Bench 2.0 (51.0), LiveCodeBench-V6, ResearchRubrics, τ²-Bench, and BrowseComp with context-manager settings. Release positioning emphasizes robust long-horizon agent execution, tool-use reliability, and cost-efficient decoding relative to larger-parameter peers, with deployment support across major open inference stacks and community integrations.